Error Mitigation in Computational Design of Sustainable Energy Materials

Rune Christensen

Research output: Book/ReportPh.D. thesisResearch

504 Downloads (Pure)

Abstract

Transportation based on sustainable energy requires an energy carrier, which is able to store the predominately electrical energy generated from sustainable sources in a high energy density form. Metal-air batteries, hydrogen and synthetic fuels are possible future energy carriers. Density functional theory calculations contribute in research and development of these technologies.

Systematic errors are present in calculations with general gradient approximation functionals for all three technologies. Such functionals will in many cases be the best compromise of computational cost and accuracy if not for the systematic errors. In this thesis it is shown how the systematic errors can be mitigated.

For different alkali and alkaline earth metal oxides, systematic errors have previously been observed. These errors are primarily caused by differences in metal element oxidation state. The systematic errors can be significantly reduced by using metal chlorides rather than pure bulk metals as point of reference for metal oxide energies.

Systematic errors in gas phase CO2 reduction reactions have previously been attributed a molecular O-C-O backbone structure. They are through error correlations found to be caused by individual C=O bonds. Energy corrections applied to C=O bonds significantly reduce systematic errors and can be extended to adsorbates.

A similar study is performed for intermediates in the oxygen evolution and oxygen reduction reactions. An identified systematic error on peroxide bonds is found to also be present in the OOH* adsorbate. However, the systematic error will almost be canceled by inclusion of van der Waals energy. The energy difference between key adsorbates is thus similar to that previously found.

Finally, a method is developed for error estimation in computationally inexpensive neural networks. The method can validate the use of a neural network for emulation of density functional theory calculations for given atomic configuration.
Original languageEnglish
PublisherDepartment of Energy Conversion and Storage, Technical University of Denmark
Number of pages210
ISBN (Print)978-87-92986-58-0
Publication statusPublished - 2017

Fingerprint

Dive into the research topics of 'Error Mitigation in Computational Design of Sustainable Energy Materials'. Together they form a unique fingerprint.

Cite this